Abstract: Long-term monitoring of grassland phenology in large-scale areas by remote sensing is an important area of ecological research and a core part of global change research. Based on the GIMMS 3g Normalized Difference Vegetation Index (NDVI) of the 30-year (1986-2015) dataset, the Savitzky-Golay filtering method was employed to reconstruct the time series. A dynamic threshold method was used to extract phenological parameters of Chinese natural grassland (start of the growing season, SOS; end of the growing season, EOS; and length of the growing season, LOS). Then, 1998 was selected as the time demarcation point to analyze the temporal trends in evolution and spatial differentiation of grassland phenological characteristics in the two time periods. It was found that: 1) The national average annual SOS, EOS, and LOS were mainly concentrated around 100-140 days, 260-290 days, and 130-170 days, respectively; 2) Phenological trends have not changed significantly over the 30 years with a range of -0.3 to 0.3 d·yr^-1; 3) Before 1998, the SOS of grassland was 0.37 d·yr^-1 earlier and the EOS was 0.43 d·yr^-1 later on average, and there was a big difference in the area of grassland. After 1998, grassland phenology began to change. The SOS of grassland showed the largest change with an average advance of 0.28 d·yr^-1; 4) The phenological characteristics of alpine subalpine meadow, mountain grassland, flat grassland, and alpine subalpine flat grassland changed in a completely opposite direction after 1998. The time trends identified in this paper are consistent with observational data and the results of other studies.

Key words: natural grassland of China, phenology, GIMMS NDVI 3g, dynamic threshold, temporal and spatial distribution characteristics